Best practices

Use one synchronized node as sensor controller (MCU)

Using one controller in each system allows a better control of the nodes' initialization and also simplifies the node synchronization. Often, the controller is the only synchronous node in the system.

Process data on the GPU as much as possible

For performance reasons, you should do data transformation on the GPU. The GPU is a lot better than the CPU to process data in parallel. Consider to copy back in CPU memory only if needed.

Use the StructuredBuffer class provided by SensorSDK to represent GPU data, and use the extension method RequestAsyncReadback to get back data on the CPU while ensuring proper synchronization with the commands queued into the command buffer.

For point cloud data on the GPU, use the PointCloud class. See Use point cloud data for more details on how to get back point cloud data on the CPU.

Don't modify samples

To simplify project upgrades, it's recommended to avoid modifying any of the sample content provided by SensorSDK, such as materials, sensors, and scenes.